# SAT Statistics: Sampling, Assignment and Causation | topin

On this page

## Ask two separate questions about every study

College Board’s [data-analysis content page](https://satsuite.collegeboard.org/sat/whats-on-the-test/math/types/problem-solving) includes inference from samples, margin of error and evaluating observational studies and experiments. Start every study question with two questions: how were participants selected, and how was any treatment allocated? A passage can use the word random in either answer. The word alone does not justify every kind of conclusion.

Our suggested scratch-paper labels are “selection” and “assignment”. Under selection, identify population, sampling frame and how participants entered. Under assignment, identify groups and how treatment was determined. If no treatment was assigned, mark observational rather than assuming a comparison is an experiment. This small separation prevents a student from treating a randomly selected survey as proof that one behaviour caused another.

The examples below use fictional studies and simple numbers. They are demonstrations of statistical reasoning, not reports of real scientific findings. A valid conclusion must fit the facts supplied by the scenario. Even when a causal experiment is described, inspect who participated before generalising. Evidence about a particular volunteer group does not automatically establish the same effect for every age, country or setting.

## Original example: a random school survey

A school has 1200 students. Researchers randomly select 120 from its complete student list and ask whether they prefer a later start time. Seventy-two say yes. The sample proportion is 72/120 = 0.60, or 60%. With an appropriate random sampling and response process, that provides evidence for estimating the preference in this school’s student population, subject to sampling uncertainty.

It does not establish that a later start time improves exam results, because no start-time treatment was assigned and no academic outcome is described. It also does not directly estimate preferences at every school in the city. The population named in the selection process is this school. Choosing a random group within one school does not make it a random sample from every school.

Now change selection: researchers ask the first 120 students entering the library. Even with the same 72 yes responses, the random-selection basis is missing. Library visitors at that time may differ from the full student population. The numerical proportion remains 60%, but its interpretation changes. A sample calculation and a defensible population conclusion are separate questions, and the latter depends on design.

## Original example: a volunteer experiment

Eighty volunteers join a study of two revision formats. Researchers randomly assign 40 to Format A and 40 to Format B, then administer the same assessment under the same conditions. Suppose Format A has a higher average. Random assignment helps isolate the treatment comparison by reducing systematic differences in how participants enter each group. In an appropriately controlled experiment, the result can support a causal comparison for those conditions.

The volunteers were not stated to be randomly sampled from all students. Therefore the design does not automatically support extending the effect to every student in the country. “Random” appears in the assignment step, not the population-selection step. A tempting answer can correctly identify causation while incorrectly adding universal generalisation. Check both parts of a compound conclusion, not only its first clause.

Now suppose volunteers choose their preferred format themselves. The average difference can reflect existing differences between groups, such as motivation or prior habits. The comparison is then observational with respect to format choice. More participants do not repair the missing random assignment. A large self-selected study can still have a confounding problem, and a small well-designed experiment can still have limited generalisability.

## Original example: combining sampling and assignment

Researchers randomly select 200 students from a district’s eligible list and randomly assign them to two permitted study routines. They measure the same outcome and maintain the same other conditions. If the design and participation are appropriate, the selection supports inference to the sampled district population and the assignment supports a causal comparison of the routines. Each conclusion comes from a different part of the design.

Even here, do not claim the result applies to everyone everywhere. The sampling population is the eligible district list, and the treatment and outcome are specific. If the study concerns a four-week routine for one subject, a claim about years of learning in every subject goes beyond the scenario. Stronger design does not remove the need to match the scope of the conclusion to the described conditions.

Our evaluation routine first checks direction: what group had what result? Then checks design: was treatment assigned? Finally checks scope: which population and circumstances are covered? An answer can fail at any stage. Avoid choosing an impressive conclusion simply because it includes phrases such as statistically significant or random. The words must describe the actual comparison and the actual process supplied in the passage.

## Original example: association with an alternative explanation

A fictional survey finds that students who use a study app have higher average marks than students who do not. The passage does not state random assignment. The finding describes association. It does not prove the app caused the difference: users might already study more, have stronger prior preparation or receive other support. Those are possible alternative explanations, not established facts about the fictional participants.

The careful conclusion is that the study alone cannot isolate the app’s causal effect. It is not that the app certainly has no effect, nor that motivation definitely caused the difference. Both extremes add evidence not supplied. A limitation statement should identify what remains unresolved rather than announce a replacement cause. This same distinction matters in [science inference questions](/articles/sat-science-inference-controls).

If researchers instead randomly assign eligible participants to use or not use the app and control relevant conditions, the causal comparison becomes stronger. Check the actual passage for these design changes. Do not import them because you expect scientists to have done them. The right answer depends on the described method, and a question can deliberately omit the condition needed for a tempting causal claim.

## Original example: interpret a margin of error

A fictional poll estimates that 58% of a population favours a proposal with a margin of error of 4 percentage points at the stated confidence level. The corresponding interval is 54% to 62%. The margin concerns uncertainty in the population estimate under the poll’s method; it does not mean every individual has an answer “within 4%” or that exactly 4% of respondents answered incorrectly.

If another sample estimates 61% with the same margin, its interval is 57% to 65%. These overlapping intervals do not by themselves prove a real increase from 58% to 61%. Do not treat the difference between two point estimates as automatically conclusive. Read the question’s requested interpretation and use only the uncertainty information supplied. The intervals provide context, not a guarantee that any particular number is the true population value.

A smaller stated margin means a narrower interval, not necessarily a less biased study. Bias in selection or measurement is different from sampling variability. Increasing sample size can reduce sampling uncertainty under a comparable appropriate method, but it does not automatically correct a systematically unrepresentative sampling frame. Keep “more precise” and “more representative” distinct when evaluating a proposed improvement.

## Use a design table, then verify transfer to fresh questions

__Original study-design map; conclusions require an appropriate implementation__
| Selection and assignment                   | Population inference                              | Causal comparison                          |
| ------------------------------------------ | ------------------------------------------------- | ------------------------------------------ |
| Random sample; random treatment assignment | Supported for the sampled population              | Supported under experimental conditions    |
| Random sample; no treatment assignment     | Can estimate that population                      | Association alone does not establish cause |
| Volunteers; random treatment assignment    | Not automatic for all students                    | Can support effect within the experiment   |
| Volunteers choose treatment                | Generalisation not established by random sampling | Causal effect not isolated by assignment   |
| Larger biased sample                       | Bias remains a concern                            | Size does not create an experiment         |

Practise by changing one design feature at a time: random selection to volunteers, random assignment to choice, district population to one school. Explain exactly which conclusion changes. Use the [data-analysis overview](/sat/math/problem-solving-and-data-analysis) for surrounding skills and the [error-log guide](/articles/sat-error-log-score-plateau) to record selection, assignment, scope or uncertainty as the missed distinction.

[topin’s free SAT mock](/sat/practice-test), marked on the official scale, can check this reasoning in mixed timed work. Do not memorise the table as a substitute for reading the scenario. The table describes design principles, while the question supplies the actual implementation. Finish each answer by identifying the sentence that supports selection, the sentence that supports assignment and the population named in the conclusion.

## Try a full SAT mock free

Timed like test day, every section scored, every answer explained.

[Take the free mock ](/sat/login)

## FAQs

What is the difference between random sampling and random assignment?

Sampling selects participants from a population; assignment allocates participants to treatment groups. They support different conclusions.

Can a randomly sampled survey prove causation?

Not from association alone. Check whether a treatment was assigned in an appropriate experiment.

Can a volunteer experiment generalise to everyone?

Not automatically. Random assignment does not establish that volunteers represent a wider population.

Does a larger sample remove selection bias?

No. More observations from a biased selection process do not by themselves make it representative.

Does a 4-point margin of error describe individual respondents?

No. It describes uncertainty around the population estimate under the stated method and confidence level.

## Sources (checked 5 October 2026)

* [College Board: Problem-Solving and Data Analysis content (checked 5 October 2026)](https://satsuite.collegeboard.org/sat/whats-on-the-test/math/types/problem-solving)
* [College Board: using the Student Question Bank (checked 5 October 2026)](https://satsuite.collegeboard.org/practice/student-question-bank)
* [College Board: Reading and Writing domains (checked 5 October 2026)](https://satsuite.collegeboard.org/sat/whats-on-the-test/reading-writing)

## Related articles

[All SAT study guides ](/sat)

* [SATStudy plans & strategySAT Math timing: a 35-minute module strategyBuild a practical SAT Math pacing plan for 22 questions in 35 minutes, with checkpoints, skip decisions and an original example of choosing the fastest method.Updated 5 Oct 2026](/articles/sat-math-module-pacing)
* [SATStudy plans & strategySAT Math word problems: equations, rates and unitsTranslate SAT word problems into equations using quantities and units, with original ticket, rate, mixture and conversion examples and common modelling traps.Updated 5 Oct 2026](/articles/sat-math-word-problems-units)
* [SATStudy plans & strategySAT quadratic parameters: discriminants and root conditionsSolve SAT quadratic parameter questions using discriminants, vertex form and root relationships, with original one-root, no-root and non-quadratic examples.Updated 5 Oct 2026](/articles/sat-quadratic-discriminant-parameters)
* [SATStudy plans & strategySAT reverse percentages and successive changesSolve SAT reverse-percentage questions with multipliers, different base amounts, successive changes and percentage-point comparisons using original examples.Updated 5 Oct 2026](/articles/sat-reverse-percentages)